Content uploaded by Walter Terkaj
Author content
All content in this area was uploaded by Walter Terkaj on Oct 23, 2017
Content may be subject to copyright.
CIRP Template v4.0
A virtual factory approach for in-situ simulation to support production and
maintenance planning
Walter Terkaj1, Tullio Tolio1,2 (1), Marcello Urgo2
1: ITIA-CNR, Institute of Industrial Technologies and Automation, National Research Council, Milano, Italy
2: Manufacturing and Production Systems Division, Department of Mechanical Engineering, Politecnico di Milano, Milano, Italy
Structured methodologies and tools for the tailored design of factories are more and more adopted by suppliers of manufacturing systems but usually
discontinued after the design phase. The use of an ontology-based virtual factory, continuously synchronised with the real plant, is proposed to
guarantee digital continuity and enable in-situ simulation during the operating phase of a factory. This digital counterpart of the system can be used for
integrated shop-floor simulations to assess future impact of production and maintenance planning decisions. An industrial application is provided in the
context of roll shops, i.e., systems devoted to the grinding of cylinders for rolling mills.
1. Introduction
The design of a manufacturing system is a complex engineering
problem requiring the joint use of multiple structured
methodologies and tools [1]. These tools, exploited during the
design phase [2], are typically discontinued as soon as the
installation of the manufacturing system enters the operation
phase. However, their residual value could be high if they were
integrated in a Virtual Factory environment to support:
the performance evaluation of a factory starting from a specific
state to compare different management decisions [3].
possible reconfigurations of the system [4]. This scenario is
becoming more and more relevant given the short product
lifecycles and, thus, the need of adapting an existing
manufacturing system to the production of new part types.
the ramp-up phase of production systems, often due to the fact
that system integrators usually evaluate the performance of a
factory design grounding on incomplete models and data
(maintenance policies, failures modes and operating conditions
are typically not available at the design phase).
The focus of this paper is on the first item in the list, proposing
the application of an ontology-based virtual factory approach to
evaluate the impact of planning and maintenance decisions
during the operation phase of a manufacturing system. This goal
can be achieved via in-situ simulation by exploiting and adapting
the set of tools available at the design phase. However, these tools
require a digital model of the factory continuously synchronised
with the real one. This demands the integration of various
monitoring and planning tools modelling and storing the
evolution of the system and its current status. The availability of
these data in a shared format and location enables the
initialization of in-situ simulation tools to rapidly assess the
impact of decisions to be taken in a short-term horizon.
The implementation of in-situ simulation entails a wide range of
scientific problems to be tackled. Firstly, a smooth model
initialization is poorly supported in almost all of the commercial
discrete simulation software tools. A second one is the integration
of multiple and heterogeneous sets of data and tools into a unique
and coherent scheme. Nevertheless, overcoming these technical
limitations opens the way to significant benefits from the
industrial point of view. In fact, the availability of rich data sets
and the use of engineering tools in the every-day life of a
production system is neither frequent nor smooth. These tools
could clearly provide a much more structured approach to take
decisions in a complex environment and, consequently, a
potential improvement of the factory performance.
Section 2 gives a review of virtual factory approaches and their
role in coupling digital and real factories for production and
maintenance management. Section 3 outlines the proposed
methodology, specifically the representation of production
system history and how multiple histories from different sources
can be merged achieving digital continuity. Section 4 provides the
framework scheme for digital continuity and in-situ simulation
applied to a reference case. Section 5 describes the
implementation in a real industrial environment and Section 6
provides a summary and outlook for further developments.
2. Literature Review
The main benefits coming from Virtual Manufacturing and
Virtual Factory consist in a multi-layered integration of the
information related to: (1) various activities along the factory and
product lifecycle; (2) hardware, software, and human resources;
(3) real world and virtual world [5]. Through this effort,
modelling and simulation approaches can be extensively used to
support and speed-up the decision process, not only in the design
stage but also for management decisions [3]. In particular, the
close relationship between production and maintenance planning
often requires these sets of decisions to be jointly taken. More
generally, the need to consider production, maintenance and the
quality from an overall point of view has been highlighted in [6].
This class of integrated approaches is typically limited by the
lack of updated information on the status of the system and could
benefit from the availability of a virtual factory model to support
them. However, the continuous upgrade, update and maintenance
of Virtual Factory models and tools along the factory lifecycle has
to face practical problems that hinder the usage of approaches
typical of the design phase also during the operations of a factory.
Indeed, simulation tools are usually developed and used by
experienced operators (mainly machine tool builders and system
integrators, given their specific expertise and frequency of
needing to solve this type of problem) and rarely transferred to
the customer (i.e. the owner of a factory). In addition, these
Contents lists available at SciVerse ScienceDirect
CIRP Annals Manufacturing Technology
Journal homepage: www.elsevier.com/locate/cirp
models are usually not updated, thus failing to guarantee the
digital continuity. Nevertheless, the realization of a full-scale
Virtual Factory is still far from being reached both on academic
and commercial sides. Enabling technologies are being
investigated by academics to fulfil the concepts sketched by the
early literature works, while large ICT market players propose
software suites that still lack full integration and/or are not
affordable for a large share of industrial companies.
The exploitation of a digital model of a factory, coupled with a
continuous synchronisation of the information coming from the
real system, has been previously addressed in [7]. The authors
proposed the use of a simulation model that can be initialized to
match the real state of the manufacturing system under study.
The authors identified the following problems to be tackled: (i)
the acquisition and validation of the input data, (ii) the
responsiveness of the analysis and (iii) the capability of creating a
snapshot of the real system to initialize the simulation model. In
particular, the first and third points highlight a data
synchronisation and consistency problem between the real and
the virtual environment.
Indeed, the generation and management of digital factory data
is a key problem of Virtual Factory approaches. Several authors
addressed the need of relying on data schemas in compliance
with existing standards like STEP [8]. The Core Manufacturing
Simulation Data (CMSD) initiative [9] proposed by NIST is one of
the efforts towards integrating the real data coming from the
shop-floor with simulation tools. Recently, the attention has been
also focused on the use of ontologies to meet the goals of
modelling, meta-modelling and interoperability [10] between the
digital tools in the Virtual Factory context to guarantee a proper
digital continuity [11][12]. The key advantages of an ontology-
based approach include (i) the exploitation of semantic web
technologies in terms of interoperability, data distribution,
extensibility of the data model, querying, and reasoning; (ii) the
re-use of general purpose software implementations for data
storage, consistency checking and knowledge inference.
The typical obstacles in the implementation of digital continuity
are highlighted in [7]: (i) the matching of the data structures of
ERP and MES databases with the data needed for the analysis, (ii)
the difficulty of coping with an enormous amount of data coming
from the real plant to get aggregate values suitable for the
analysis. These key problems can be addressed by adopting an
ontology-based approach [12] to support the factory design
phase. Herein, this approach is further developed by extending
the ontological data model to represent the continuous evolution
of factory objects during the operation phase of a plant. The need
of acting at different levels of detail is also considered and
managed through a multi-scale representation of the data.
3. Modelling the history of factory objects
A unifying modelling of the evolution of the objects in a factory
(i.e. products, processes and production resources) has to cope
with a wide range of heterogeneous data and levels of detail.
These streams of data (histories) come from different sources,
e.g., monitoring systems (i.e. a real history), production-planning
methods (i.e. a planned history), performance evaluation tools
(i.e. a simulated history). Data relate to both physical
characteristics (e.g. the placement of an object) and abstract
properties (e.g. state transitions) and the dynamic evolution of
the factory objects must be coherent with their static
representation. To meet these requirements, an OWL (Web
Ontology Language) ontology data model [13], based on the
standard Industry Foundation Classes (IFC), has been extended
with additional classes, as sketched in Figure 1, where the novel
classes have a grey background.
Figure 1. Simplified fragment of the ontology-based data model. Boxes
represent OWL classes, whereas arcs represent property restrictions
(universal quantification).
The generic factory object (FactoryObject class) is characterized
by its placement, geometrical representation and property sets,
but also by its finite state machine. Each state (ObjectState class)
composing the finite state machine can be associated with a
specific representation of the object. According to the different
specialization of the FactoryObject (e.g. machine tools,
transporters, workpieces, etc.), specific subclasses of the
ObjectState class are provided. For instance, a machine tool will
be linked to instances belonging to classes representing the idle,
working and failed states.
A FactoryObject can be linked to as many as necessary instances
of FactoryObjectHistory to describe its history. A piece of history
can be decomposed both time-wise (i.e. defining sub-intervals)
and hierarchically (i.e. decomposed into the history of its object
components). The first option entails the capability of formalizing
detailed and specific history data coming from a real monitoring
system, while the second allows to model the history of an object
aggregating the histories of its components (e.g., a workstation is
failed if some of its components are failed).
A FactoryObjectHistory is characterized by the start and end
time of the corresponding time interval. However, a detailed
record of the evolution of factory objects along the time is not
always available. Data coming from performance evaluation
approaches (e.g., mathematical methods) typically provide a
performance measure only in terms of aggregate indicators (e.g.
mean and variance). To manage these heterogeneous cases, since
the factory object behaviour can be described by a set of possible
states, the class StateFrequency was introduced to define the
fraction of the time interval spent in a specific state together with
other statistics. This modelling pattern applies to both aggregate
intervals (where more than one state may have a frequency
greater than zero) and detailed intervals (where only one state
has a frequency equal to 100% in the considered time span).
Finally, the history of a product can be associated with a
placement in the space, to represent movements and trajectories
with the desired level of detail. The placement can also be defined
in relation to another object (e.g., a workpiece on an AGV
transporter), linking histories of different objects and providing a
concise representation for the routing.
Also the class FactoryObjectHistory can be specialized according
to the characteristics of the corresponding factory object, e.g., a
history interval of a buffer is characterized by the buffer level
and/or by a descriptive statistics representing the observations
of the buffer level during that time interval. Figure 1 shows the
links between object state and object history to unify data coming
from different sources and related to different time spans in a
single representation of the production system evolution.
FactoryObject
ObjectState
FactoryObject
History
StateFrequency
hasStateMachine hasStateFrequency
hasState
Representation
hasObjectPlacement
hasStateRepresentation
Placement
hasRepresentations
hasPlacement
hasHistory
RatioMeasure
hasStayRatio
PropertySet
hasPropertySet
4. Digital continuity and in-situ simulation
The history model of factory objects can be exploited to
guarantee the digital continuity between the real factory and its
virtual model. Historical data can be collected and stored in a
distributed way, while keeping an overall coherence thanks to the
virtual factory model. This entails the modelling of the evolution
of a system in time by joining together portions of the history
generated by different sources, e.g. the monitoring of the physical
system, a planned set of events (e.g., a production plan) or
forecasts of external factors (see Figure 2).
Figure 2. Framework for digital continuity and in-situ simulation.
This schema paves the way to in-situ simulation approaches,
through the seamless integration of simulation tools and the real
environment. As shown in Figure 2, a snapshot of the current
state of the system (condensed in a input history), together with a
set of planned and forecasted histories, can be used to feed a
simulation model and get a simulated history of the evolution of
the system (output history). This approach provides the capability
of a fast assessment of the impact of decisions grounding on the
current state of the whole factory and is specifically relevant for
short-term management decisions, e.g., production planning and
maintenance planning.
Depending on the envisioned application, different data sources
must be elaborated and aggregated in the shared semantic
repository. A highly detailed history is necessary for all the
physical factory objects if the goal is to run a virtual reality
animation of the factory. On the contrary, only the most recent
history of the system is necessary if the goal is to warm-start a
Discrete Event Simulation (DES). Herein, the latter scenario is
considered, aiming at demonstrating the use of this approach to
support production and maintenance planning in a flow line with
five process steps processing a single part type (see Figure 3).
The first and last stations are manual, whereas the other ones are
automatic. The third station consists of two parallel machines.
The flow line is balanced and the automatic stations are
characterized also by their failure modes. Inter-operational
buffers separate the stations.
A Virtual Factory model is implemented for the considered flow
line together with a configurable DES model. Since the connection
with a real plant is not available in this case, the data generated
by a previous simulation are used to mimic the behaviour of the
real plant. The input history of the system consists in the
snapshot of the objects in it (i.e. position of the parts in-progress;
status of the automatic stations) and event-related information
about the recent past (i.e. time when the parts in-progress
entered the buffer/workstation where they are placed). The goal
of in-situ simulation is assessing the impact of management
decisions, specifically different maintenance and loading policies,
on the short-term performance of the production line. The
automatic stations must undergo a preventive maintenance
whose duration is three hours. Four different schedules are
considered for the maintenance operations (1, 2, 3, 4) and two
policies (A, B) defining the workload balancing for the two
parallel machines in the third station of the line. External factors
are taken into consideration in terms of parts arriving from the
previous manufacturing stage.
Figure 3. 3D virtual reality representation of the test case.
The combined maintenance and loading policies were
simulated using 40 replicates, 24-hour long, initializing the
simulation model with two different input histories (h1, h2)
representing different states in the system in terms of number of
in-process parts, their location in the system, and state of the
machines. Arena by Rockwell Automation was used but, since it
does not provide a built-in warm-start function, the initialization
of the model was done through a specific software connector
presented in [13]. The response of the experiments is the
throughput (TH) of the flow line measured in parts per day.
Figure 4. Boxplot of the throughput of the flow line.
Figure 4 shows the boxplot of TH. Starting from status h1, the
highest average TH is given by the combined policy 3-A (i.e.
maintenance policy 3, loading policy A), whose TH is 32% higher
than the policy with the lowest average TH (i.e. policy 2-A).
However, it can be noticed that the policy 3-A has a higher
variance compared to other ones (e.g. policy 1-B) and the best
solution should be chosen taking in consideration a trade-off
between the average TH and its variance. The maintenance policy
influences TH more than the loading policy since it is not possible
to implement very different policies in a flow line with only one
parallel station. Starting from h2 a rank reversal of the policies
takes place: policies 1-B and 4-A show the highest average TH
with a low variance as well. Policy 3-A, performing well with
input history h1, is even worse than policies 1-A and 4-B.
Hence, the initial status of the system can significantly impact
on the performance of a management policy and the proposed
in-situ simulation approach can support the selection of proper
management decisions accordingly.
Finally, in-situ simulation entails a paradigm change since the
traditional warm-up identification and elimination phase is
replaced by the warm-start initialization of the model. Short
simulation runs are executed to assess the impact of a decision in
the short term and statistical relevance is achieved through
multiple replicates. Referring to the previous experiments, each
simulation run took less than five seconds, thus confirming the
viability of the in-situ simulation approach as a support for taking
management decisions.
5. Application case: roll shop
The application case relates to the production and maintenance
planning of a roll shop, devoted to the grinding of cylinders for
rolling mills. Exhausted cylinders, whose surface has been
damaged during the rolling process, are changed frequently (from
two hours up to 30 minutes). Cylinders are cooled (only for hot
rolling process) and the bearings could be taken apart before the
grinding. Different machines are used for the different class of
cylinders (working, intermediate and backup rolls) according to
their dimensions. Grinding machines for backup rolls have a
higher power and can also process other types of cylinders after a
setup. Cylinders, weighing from 10 to 100 tons, are moved using
overhead and semi-gantry cranes that are significantly stressed
and require frequent inspection and preventive maintenance
(wire rope inspection and substitution, hook inspection, etc.),
keeping the crane inactive for about a week. Hence, maintenance
and production must be carefully planned, taking into
consideration the stock of processed cylinders, the status of the
system and the replacement schedule of the rolling mill.
Figure 5. A 3D virtual reality representation of the roll shop.
A Virtual Factory approach was implemented to support the
integrated design of a roll shop defining the layout in a virtual
reality environment (Figure 5) and evaluating its performance
through a customizable DES model. In relation to Figure 2, the
physical system is the roll shop whose simulation model is
initialized with data automatically acquired from the shared data
repository. Management decisions address the workload
assignment to the machines, while external factors are the
schedule of the replacement of cylinders in the mill. A static
schedule is adopted, even if a different simulator could be used to
model the rolling mill, jointly running multiple interacting
simulation models [14]. In the specific case, the roll shop serves
two rolling mills: a tandem mill and a steckel mill. The adoption of
a maintenance policy (M1), operating opportunistic maintenance
in relation to the replacement schedule of the mill, is analysed
against the one currently used (M0). Also a loading policy (S1),
tuning the priority rules and the setup plan of the machines
according to the scheduled stops of the mill, was considered
against the old one (S0). The combinations (M0, S0) and (M1, S1)
were evaluated through the in-situ simulation approach taking as
starting point the status of machines and transporters, the
position of cylinders and cranes, the scheduled setup of the
machines. The analysis was useful to suggest the adoption of the
policy (M1, S1), guaranteeing service levels comparable with the
others but a lower average flow time of the cylinders (-1.5% for
tandem, -3.1% for steckel rolls) as shown in Table 1, providing an
earlier availability in case of unexpected replacement (damaged
cylinders, opportunistic replacement, etc.).
Table 1
Results for the roll shop application case.
Policy
(M0, S0)
(M1, S1)
BUR Roll Type
Tandem
Steckel
Tandem
Steckel
N. in the system
16
8
16
8
Avg Flow Time [h]
919.38
1280.27
905.40
1240.74
Min Flow Time [h]
458.12
1141.86
502.04
1144.67
Max Flow Time [h]
1202.97
1450.21
1209.43
1342.33
6. Conclusions
An ontology-based virtual factory model, synchronised with the
real plant, was considered to enable in-situ simulation in support
of maintenance and production planning and tested in two
application cases. Further developments will address the lack of
built-in warm-start functionality in commercial DES software
packages, the difficulty in modelling and simulating real
management policies and exploiting the approach to also tackle
other decisions related to the ramp-up phase, e.g. aiming at
prioritizing a set of possible actions, or to the reconfiguration
phase, e.g. scheduling the list of reconfigurations,
Acknowledgements
Thanks to TENOVA Pomini for supporting the industrial case.
This research was supported by the EU project “RobustPlaNet -
Shock-robust Design of Plants and their Supply Chain Networks”,
Grant 609087 and by “Smart Manufacturing 2020” of the “Cluster
Tecnologico Nazionale Fabbrica Intelligente”.
References
[1] Tolio, T., Ceglarek, D., Elmaraghy, H. A., Fischer, A., Hu, S. J., Laperriere, L.,
Newman, S. T. and Vancza, J., 2010, SPECIES-Co-evolution of products, processes and
production systems. CIRP Annals - Manufacturing Technology, 59(2):672-693.
[2] Yang, X., Malak, R.C., Lauer, C., Weidig, C., Hagen, H., Hamann, B., Aurich, J.C.,
Kreylos, O., 2015, Manufacturing system design with virtual factory tools ,
International Journal of Computer Integrated Manufact uring, 28(1):25-40.
[3] Lin, M-H, Fu, L-C., 20 01, A virtual factory based approach to on-line simulation
and scheduling for an FMS and a case study, Journal of Intelligent Manufacturing,
12(3):269-279.
[4] Tolio, T., Sacco, M., Terkaj, W, Urgo, M, 2013, Virtual Factory: An Integrated
Framework for Manufacturing Systems Design and Analysis, Procedia CIRP 7:25-30.
[5] Iwata, K., Onosato, M., Teramoto, K., Osaki, S., 1997, A Modelling and Simulation
Architecture for Virtual Manufacturing Systems, CIRP Annals - Manufacturing
Technology, 44(1):399–402.
[6] Colledani, M., Tolio, T., Fischer, A., Iung, B., Lanza, G., Schmitt, R., Váncza, J., 2014,
Design and management of manufacturing systems for production quality, CIRP
Annals - Manufacturing Technology, 63(2):773–796.
[7] Kádár, B. , Lengyel, A., Monostori, L., Suginishi, Y., Pfeiffer, A., Nonaka, Y., 2010,
Enhanced control of complex production structures by tight coupling of the digital
and the physical worlds, CIRP Annals - Manufacturi ng Technology, 59(1):437–440.
[8] Newman, S.T., Nassehi, A., 2007, Universal Manufacturing Platform for CNC
Machining, CIRP Annals – Mannufacturing Technology, 56(1):459-462.
[9] Bloomfield, R. , Mazhari, E., Hawkins, J., Son, Y-J, 2012, Interoperability of
manufacturing applications using the Core Manufacturing Simulation Data (CMSD)
standard information model, Computers & Industrial Engi neering, 62(4):1065-1079.
[10] Ciocoiu, M., Nau, D., Gruninger, M., 2001, Ontologies for Integrating Engineering
Applications, J. Comput. Info. Sci. Eng., 1:12-2 2.
[11] Agyapong-Kodua, K., Lohse, N., Darlington, R., Ratchev, S., 2013, Review of
semantic modelling technologies in support of virtual factory design, International
Journal of Production Research, 51(14):4388-4404.
[12] Kádár, B., Terkaj, W., Sacco, M., 2013, Semantic Virtual Factory supporting
interoperable modelling and evaluation of production systems. CIRP Annals -
Manufacturing Technology, 62(1):443-446.
[13] Terkaj, W., Urgo, M., 2014, Ontology-based Modeling of Production Systems for
Design and Performance Evaluation. Proceedings of the 12th IEEE INDIN, 748-753.
[14] Pedrielli, G. , Sacco, M., Terkaj, W., Tolio, T., 2012, An HLA-based distribute d
simulation for networked manufacturing systems anal ysis, Journal of Simulation,
6(4):237-252.
